C/EBP Family Member Gene Fusions and Cooperating Genomic Events Define Two Distinct Molecular Sub-Groups in Adult B Precursor ALL

CCAAT-enhancer-binding proteins (C/EBP) are important regulators of myeloid differentiation and cell cycle control. The C/EBP family member CEBPE has been implicated in B precursor acute lymphoblastic leukemia (BCP-ALL) through IGH locus gene fusions and rare germline single nucleotide variants. ‘ZEB2 (p.H1038R)/IGH::CEBPE’ BCP-ALL is a provisional diagnostic entity in the current ICC classification. However, precise diagnostic definitions for this subtype are lacking.

We analyzed n=2,845 BCP-ALL transcriptomic profiles of own (GMALL: n=571; MLL: n=286) and published cohorts (St Jude: n=1,988), including pediatric (n=1,285) and adult cases (n=1,560). Of these, we identified n=43 BCP-ALL patients with either C/EBP family member gene fusions or corresponding gene expression signatures and absence of established subtype definition. To define the molecular profile of C/EBP BCP-ALL, we performed transcriptomic (RNA-Seq, n=31) genomic (WGS / WES: n=16; DNA EuroClonality capture analysis: n=23; SNP arrays: n=18) and epigenomic (DNA methylation arrays: n=17) profiling.

Unsupervised data analysis of the largest sub-cohort (n=25; GMALL) identified two distinct gene expression clusters – C1 (n=13) and C2 (n=12) – which shared overlapping gene expression signatures, separating C/EBP candidates from remaining BCP-ALL subtypes and from each other (C1 vs. C2). Robustness of this cluster separation was confirmed on gene expression data of independent cohorts (MLL: n=8; St Jude: n=10) and by unsupervised DNA methylation analysis of GMALL data.

Gene fusions involving C/EBP family members were clearly enriched in the novel gene expression clusters (n=36/43 vs. n=3/2,813 in remaining BCP-ALL; p<0.001) with n=7 cases sharing the gene expression profile without a C/EBP driver fusion call. CEBPE fusions were specific for cluster C1 (C1: n=11/25 vs. C2: n=0/18, p<0.001), while IGH::CEBPA (C1: n=8/25 vs. C2: n=11/18; n.s.) and IGH::CEBPB (n=5) or IGH::CEBPD (n=1)fusions occurred in both clusters. IGH locus DNA capture analysis identified in n=6/25 GMALL cases C/EBP fusions which were missed by RNA-Seq. Genomic profiling revealed subtype-specific cooperating events. Cluster C1 was characterized by ZEB2 p.H1038R (84%), NRAS/KRAS activating (74%) single nucleotide variants and CDKN2A deletions (88%; p<0.01 for all comparisons to C2 and to remaining BCP-ALL). C2 showed a less distinctive profile with enrichment of FLT3 activating SNVs (28%; p=0.01 for comparison to remaining BCP-ALL) and a tendency for IKZF1 deletions (42%; n.s.). Notably, ZEB2 p.H1038R was also observed in n=2 C2 cluster cases and in n=10 C cluster cases not harboring a CEBPE gene fusion, indicating that our gene expression clusters extend beyond the suggested genomic definition of ZEB2 (p.H1038R)/IGH::CEBPE. Comparison to normal B lymphopoiesis showed highest proximity of cluster C2 to pre-B I cells whereas C1 was closest to pre-B II cells, suggesting different developmental trajectories of C/EBP ALL subtypes.

C/EBP ALL was mainly observed in adults (median age: 46 years, range 13-88 years). A total of n=22 outcome-evaluable C/EBP ALL patients (median age: 49 years, range 18-70 years) were treated on GMALL protocols. One patient died during induction therapy. Complete MRD negativity in n=16/19 (84%) patients with MRD measurement after 1^st consolidation indicated an overall favorable therapy response of C/EBP ALL. However, only n=15/22 (68%) patients achieved long-term complete remissions, including cases with allogenic stem cell transplantation in 1^st CR (n=2). In total, n=8/22 (36%) patients experienced mostly late relapses with only n=3 patients achieving a durable 2^nd complete remission after salvage. Despite favorable initial treatment response, C/EBP ALL patients might harbor an increased risk of late relapse, possibly related to their overall advanced age. Both clusters, C1 and C2 shared this clinical phenotype despite their distinct molecular background.

To facilitate diagnostic identification of C/EBP clusters C1 and C2, we trained a machine learning classifier on the GMALL gene expression data set which separated these from remaining BCP-ALL with accuracies of 99.7% in training data and 99.0% (MLL) or 99.4% (St Jude) in independent hold-out cohorts.

Integration of genomic, epigenomic and transcriptomic data defined two molecular distinct adult BCP-ALL subtypes harboring C/EBP family gene fusions.